Removing video cross-feed from the audio mix

July 15, 2007 4:58 PM

At a recent lecture, I taped from the soundboard & from some microphones. I meant to record these to separate channels, but I accidentally left it in low quality mode - mono, 22khz. It sounds like video got thrown into the mix. Can I salvage the audio?

This three-minute excerpt is all I really need. It's from a backup recording device, & this is the only gap from the main recording.

Want a quick listen? Here's an mp3 (1.4 meg), with frequencies below ~100hz & above ~5khz removed to try to be nicer to speakers.

If you're brave, here's the original, uncompressed, excerpt. (8 meg wav file)

posted by Pronoiac to computers & internet (9 comments total)

Short answer: no.
posted by pompomtom at 5:34 PM on July 15, 2007

The hum is definitely removable, but it doesn't sound like there's enough signal regarding the speech itself to actually recover.
posted by effugas at 5:52 PM on July 15, 2007

maybe. doubtful.

If you have the original video, you could load the sound into Audacity and sync it up exactly with the recording. Then follow these instructions on removing lyrics from songs. (Hint: think of the video track as the lower/right channel) It might not work at all, maybe will work good enough.

----
on preview and listening to the mp3, I guess you really mean the video got mixed in. If you can replicate the buzzing, give it a try, but I don't think there's enough original audio of the speaker to work even if the buzzing goes away.
posted by niles at 5:55 PM on July 15, 2007

Yes. There used to be a feature in Cool Edit for removing tape hiss that'll do the trick. Essentially, you take a sample of the noise then apply it to the entire recording. The feature is "Noise Removal" (under Effects) in Audacity. Be warned that it'll likely leave you with some weird phasing in the resulting audio.
posted by jdfan at 5:59 PM on July 15, 2007

So.. Here is the power spectral density (PSD) using the Welch method.

For reference, PSD, Welch.

So, no. At least I don't think so. The noise is spread across far too many frequencies to filter.
posted by Chuckles at 6:32 PM on July 15, 2007

Actually, the videocamera was the primary; it sounded decent. This noisy bit is covering when I swapped tapes.

I'll double check noise removal. It looked like it backed up what Chuckles wrote: the buzz spread too widely to be isolated.

The video buzz is coming from the computer slide show presentation, I think. Also, the recording this is from was recording with auto-gain turned on, so the signal changes from beat to beat. So I doubt I can replicate the buzz exactly. I might be able to pin it down a bit, though.
posted by Pronoiac at 7:03 PM on July 15, 2007

I've done several audio restoration projects in the past, so I figured I may as well see what I could do with it in 10 minutes. I got it to the point where it was coherent enough to be used as wiretap evidence in court, but nothing beyond that.

At best, were a good amount of time devoted, you could get the speech clear enough that someone with a good ear could make out the words well enough to transcribe it. To most, it would probably sound like Charlie Brown's teacher.

If you're curious, I can email you the quick cleanup I did to the file.
posted by yorick at 9:41 AM on July 16, 2007

Yorick: Yes, I'm interested in the quick cleanup.

Hmmm... Subtitles would work...

As might using Charlie Brown's teacher, as a gag.
posted by Pronoiac at 11:49 AM on July 16, 2007

Looking back at this: I looked at converting the audio to an image - like the reverse of the Aphex Twin face embedded in the audio - then filtering out noise with Photoshop.

The verdict: Unsalvageable. (Within my time & money & energy constraints.)
posted by Pronoiac at 4:11 PM on March 18, 2008